| from PIL import Image | |
| import numpy as np | |
| import pandas as pd | |
| from datasets import Dataset | |
| import pyarrow as pa | |
| import pyarrow.parquet as pq | |
| # 读取图片并转换为 numpy 数组 | |
| img_path = "lena.png" | |
| img = Image.open(img_path) | |
| img_array = np.array(img) | |
| # 将图像数组转换为 DataFrame 格式 | |
| df = pd.DataFrame(img_array.reshape(-1, img_array.shape[2]), columns=[f'pixel_{i}' for i in range(img_array.shape[2])]) | |
| # 插入前几行为空的行（例如前5行） | |
| num_empty_rows = 5 | |
| empty_df = pd.DataFrame([[None]*df.shape[1]] * num_empty_rows, columns=df.columns) | |
| df = pd.concat([empty_df, df], ignore_index=True) | |
| # 将 DataFrame 转换为 Dataset 格式 | |
| dataset = Dataset.from_pandas(df) | |
| # 将 Dataset 转换为 Parquet 格式并保存 | |
| table = pa.Table.from_pandas(df) | |
| pq.write_table(table, 'lena.parquet') | |
| print("Parquet 文件已生成：lena.parquet") | |